Interprets conditions
adapts- Situated judgment
- Detection and coordination
- Verification and intervention
- Recovery under variation
Human, AI, and Organizational Performance
AI is already performing in the work. Is control designed to keep up?
HAOP is the work-system design framework for locating how human, AI, and organizational performers jointly produce work.
It shows where verification, grounding, authority, and correction must be designed before representation becomes consequence. The work system—not the AI product—is the unit of success and failure.
Framework under active development · not yet empirically validated
The work system
Control path
Humaninterprets, adapts, verifies, intervenes
AIselects, transforms, routes, recommends, executes
Organizationallocates functions and authors conditions
Representationwhat the workflow presents as true
Verification gateContact with realityevidence, anchors, criteria, authority
Consequencewhat becomes harder to correct
Integrated frameworkRevision 5.0 · Deposited9 August 2026 · DOI 10.5281/zenodo.21868036
Applied methodARECC Version 1.0 · Deposited9 August 2026 · DOI 10.5281/zenodo.21868086
Current workWorkbook and ADSDSIn development · not presented as validated
The three-performer architecture
Performance is produced by the arrangement, not one actor.
HAOP extends Human and Organizational Performance into work systems where AI materially shapes what is seen, decided, or done. AI does not need autonomy or intention to operate at performer level. It needs a delegated function whose influence is not effectively contained before consequence.
The performers are analytically parallel, but they are not morally equivalent. Humans are rights-holders. AI is an artifact without moral agency. The organization holds power over the conditions of work.
Human control is not a label. It must be designed.
Performer contributions
What each performer contributes to the work system
Acts on representation
optimizes- Selection and classification
- Generation and transformation
- Ranking, routing, and suppression
- Execution within permissions and constraints
Authors conditions
contains and governs- Function allocation
- Staffing, evidence, and throughput
- Permissions, incentives, and interfaces
- Escalation, authority, and acceptable tradeoffs
The five-hazard register
Locate the condition before it becomes the event.
The five classes are a baseline operational register within HAOP's scope. They are connected, not mutually exclusive, and not exhaustive of AI risk.
01Physical interaction and machine agency
Embodied AI carries energy, mass, speed, force, and software-defined or adaptive action into physical work.
02Psychosocial conditions
Surveillance, scoring, pacing, scheduling, and machine-generated attribution can reduce autonomy, intensify work, suppress reporting, and make power continuous.
03Human performance and verification
AI can reshape attention, situational awareness, competence, judgment, alarm burden, and verification demand. Verification Overrun appears here, but the condition is authored through work-system design.
04Recursive information degradation
Representations are repeatedly summarized, classified, merged, selected, stored, retrieved, and transformed until provenance, context, uncertainty, low-frequency conditions, or situated knowledge disappears.
05Fragmented control and concentrated accountability
Consequential control fragments across vendors, procurement, configuration, leadership, and local operation while verification and accountability concentrate at the visible point of use.
From human-in-the-loop to human-in-the-design
The person at the end is not proof of control.
The work sets the demand for verification. The workflow supplies the capacity. A person can be present and still lack the knowledge, evidence, time, or authority the check requires.
01
Work flows
Human, AI, and organizational performers select, transform, route, suppress, approve, or act.
02
Position changes
A consequential transition approaches: after passage, correction becomes materially harder.
03
Demand is set
The verification function defines what must be established under the operating condition.
04
Capacity is tested
Knowledge, evidence, time, and authority must all satisfy the requirement.
05
Passage is conditional
A functional gate can proceed, return, revise, reject, pause, defer, constrain, or escalate.
The HAOP method
Purpose before technology.
ARECC—Anticipate, Recognize, Evaluate, Control, and Confirm—is inherited from industrial hygiene. HAOP's deposited contribution is its translation across the three performers and the verification architecture.
A
Anticipate
Establish True Function and unacceptable outcomes before selecting a product.
R
Recognize
Reconstruct work-as-done, decompose functions, and locate consequential transitions.
E
Evaluate
Test verification capacity, grounding, Anchor Access, reversibility, and alternatives.
C
Control
Allocate functions and engineer gates, evidence, constraints, pause, and handback.
C
Confirm
Deploy, learn, monitor, and revalidate as the work and operating conditions change.
Tools and research
Use the framework without mistaking a tool for a finding.
Each resource carries its publication or development status. The live diagnostic is an entry tool, not a safety certification and not the complete HAOP method.
True Function Diagnostic
Run the nine-question entry diagnostic across one bounded workflow. Record evidence and unknowns without producing a maturity score.
Integrated framework
The controlling source for the three performers, five hazards, Accountability by Control, grounding, verification, and the applied method.
Workbook and ADSDS
The workbook and complete AI Deployment Safety Data Sheet structure remain in development. The instrument's name and bidirectional purpose are deposited.
Pilot it. Challenge it.
Run HAOP against real work.
Bring a bounded workflow, the people who perform it, and the evidence needed to test where the framework holds, where it breaks, and what the work design must supply.